GPT-5.6 Terra

GPT-5.6 Terra

Quick answer: GPT-5.6 Terra is OpenAI's balanced mid-tier model in the GPT-5.6 family, released July 9, 2026. It achieves 96.5% on ARC-AGI and 50.4% on Agent's Last Exam at half the price of Sol — making it the recommended default for most professional and agentic workloads. Priced at $2.50 input / $15 output per 1M tokens with a 1.1M-token context window.

At a Glance

Where GPT-5.6 Terra leads

• 96.5% ARC-AGI — near-Sol abstract reasoning at half the cost
• 50.4% Agent's Last Exam — strong multi-step agentic performance
• AA Intelligence Index 55, Coding Agent Index 77.4 — highly capable for everyday professional tasks
• Best price-performance ratio in the GPT-5.6 family for most workflows
• 1.1M-token context window — same as Sol

Where it lags

• Below Sol on the highest-difficulty reasoning and cybersecurity tasks
• No ultra multi-agent mode (Sol-exclusive)
• Closed API model with no open weights

Best for: Professional developers, enterprise automation pipelines, and agentic workflows where strong capability is required but Sol's premium cost is not justified.

What GPT-5.6 Terra Is

GPT-5.6 Terra occupies the middle tier of OpenAI's GPT-5.6 family, positioned between Sol (maximum capability) and Luna (lowest cost). The "Terra" name is a permanent capability band — as OpenAI improves the GPT-5.6 generation, Terra advances alongside Sol and Luna, maintaining its role as the balanced default.

For most teams, Terra is the natural starting point. Its benchmark results sit very close to Sol's on the tasks that matter most for general agent development: 96.5% on ARC-AGI (vs Sol's 97.5%) and 50.4% on Agent's Last Exam (vs Sol's 52.7%). That ~2-point gap costs 2× in per-token pricing. On tasks that don't saturate the model's capability ceiling — routine coding assistance, document analysis, multi-turn dialogue — Terra delivers essentially equivalent output to Sol.

Terra shares the same 1.1M-token context window as Sol, making it equally capable for long-context retrieval, whole-codebase analysis, and multi-document synthesis. The primary trade-off vs Sol is access to extended reasoning modes: Sol's max and ultra effort levels are not available on Terra.

Specifications

Pricing

Cache writes are billed at 1.25× the uncached input rate; cache reads receive a 90% discount. Pricing per OpenAI pricing page as of July 2026.

Context Window

GPT-5.6 Terra has a ~1.1M-token context window — roughly 800–900 pages of text in a single request. This matches Sol's context capacity exactly, making Terra suitable for the same long-context workloads (full codebases, long contracts, multi-session analysis) at a significantly lower cost per run.

Public Benchmark Scores

Per-variant scores currently available for ARC-AGI and Agent's Last Exam. Shared family scores (GPQA Diamond, TerminalBench, SWE-Bench Pro, etc.) are reported for GPT-5.6 Sol — see the GPT-5.6 Sol page for the full benchmark table.

GPT-5.6 Terra vs Alternatives

Terra is the sweet spot in the GPT-5.6 family for most production workloads: it closes to within ~2 points of Sol on both major benchmarks while costing half as much. The gap between Terra and Luna, by contrast, is larger (8.5 points on ARC-AGI), making Terra the better default over Luna for any task with meaningful reasoning demands.

How GPT-5.6 Terra Performs on Real Agent Tasks

Terra's benchmark profile indicates strong general-purpose capability with only marginal degradation versus Sol. The 96.5% ARC-AGI score confirms it handles novel problem-solving at near-frontier level, while 50.4% on Agent's Last Exam (Benchgen's multi-step agentic evaluation) demonstrates reliable multi-tool and long-horizon task execution.

For agent builders, Terra is the recommended default tier when moving from prototype to production: capable enough to handle the majority of real-world task types, priced for sustainable at-scale operation, and sharing Sol's full 1.1M-token context window for long-document and codebase workflows. Escalate to Sol only when your evaluation data shows a consistent capability gap on specific high-stakes tasks.

Use GPT-5.6 Terra via API

from openai import OpenAI

client = OpenAI(api_key="YOUR_API_KEY")

response = client.chat.completions.create(
    model="gpt-5.6-terra",
    messages=[{"role": "user", "content": "Review this pull request and flag any issues..."}],
)
print(response.choices[0].message.content)

Related Models & Benchmarks

• Compare: GPT-5.6 Sol · GPT-5.6 Luna · GPT-5.5
• Benchmarks: ARC-AGI · Agent's Last Exam · GPQA Diamond · SWE-Bench Pro

Frequently Asked Questions

Specs and scores sourced from OpenAI's official GPT-5.6 announcement (July 9, 2026) and Benchgen evaluations; third-party benchmark scores attributed inline. Pricing cited to the OpenAI pricing page. Last updated 2026-07-21.